4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user *filename, int flags, int *empty)
127 struct filename *result;
131 result = audit_reusename(filename);
135 result = __getname();
136 if (unlikely(!result))
137 return ERR_PTR(-ENOMEM);
140 * First, try to embed the struct filename inside the names_cache
143 kname = (char *)result->iname;
144 result->name = kname;
146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
147 if (unlikely(len < 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len == EMBEDDED_NAME_MAX)) {
159 const size_t size = offsetof(struct filename, iname[1]);
160 kname = (char *)result;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result = kzalloc(size, GFP_KERNEL);
168 if (unlikely(!result)) {
170 return ERR_PTR(-ENOMEM);
172 result->name = kname;
173 len = strncpy_from_user(kname, filename, PATH_MAX);
174 if (unlikely(len < 0)) {
179 if (unlikely(len == PATH_MAX)) {
182 return ERR_PTR(-ENAMETOOLONG);
187 /* The empty path is special. */
188 if (unlikely(!len)) {
191 if (!(flags & LOOKUP_EMPTY)) {
193 return ERR_PTR(-ENOENT);
197 result->uptr = filename;
198 result->aname = NULL;
199 audit_getname(result);
204 getname(const char __user * filename)
206 return getname_flags(filename, 0, NULL);
210 getname_kernel(const char * filename)
212 struct filename *result;
213 int len = strlen(filename) + 1;
215 result = __getname();
216 if (unlikely(!result))
217 return ERR_PTR(-ENOMEM);
219 if (len <= EMBEDDED_NAME_MAX) {
220 result->name = (char *)result->iname;
221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
227 return ERR_PTR(-ENOMEM);
229 tmp->name = (char *)result;
233 return ERR_PTR(-ENAMETOOLONG);
235 memcpy((char *)result->name, filename, len);
237 result->aname = NULL;
239 audit_getname(result);
244 void putname(struct filename *name)
246 BUG_ON(name->refcnt <= 0);
248 if (--name->refcnt > 0)
251 if (name->name != name->iname) {
252 __putname(name->name);
258 static int check_acl(struct inode *inode, int mask)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl *acl;
263 if (mask & MAY_NOT_BLOCK) {
264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl == ACL_NOT_CACHED)
270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273 acl = get_acl(inode, ACL_TYPE_ACCESS);
277 int error = posix_acl_permission(inode, acl, mask);
278 posix_acl_release(acl);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode *inode, int mask)
291 unsigned int mode = inode->i_mode;
293 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
297 int error = check_acl(inode, mask);
298 if (error != -EAGAIN)
302 if (in_group_p(inode->i_gid))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode *inode, int mask)
333 * Do the basic permission checks.
335 ret = acl_permission_check(inode, mask);
339 if (S_ISDIR(inode->i_mode)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
343 if (!(mask & MAY_WRITE))
344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
359 * Searching includes executable on directories, else just read.
361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
362 if (mask == MAY_READ)
363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
368 EXPORT_SYMBOL(generic_permission);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode *inode, int mask)
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(inode->i_op->permission))
380 return inode->i_op->permission(inode, mask);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode->i_lock);
384 inode->i_opflags |= IOP_FASTPERM;
385 spin_unlock(&inode->i_lock);
387 return generic_permission(inode, mask);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode *inode, int mask)
406 if (unlikely(mask & MAY_WRITE)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode))
414 retval = do_inode_permission(inode, mask);
418 retval = devcgroup_inode_permission(inode, mask);
422 return security_inode_permission(inode, mask);
424 EXPORT_SYMBOL(__inode_permission);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436 if (unlikely(mask & MAY_WRITE)) {
437 umode_t mode = inode->i_mode;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb->s_flags & MS_RDONLY) &&
441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode *inode, int mask)
462 retval = sb_permission(inode->i_sb, inode, mask);
465 return __inode_permission(inode, mask);
467 EXPORT_SYMBOL(inode_permission);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path *path)
480 EXPORT_SYMBOL(path_get);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path *path)
493 EXPORT_SYMBOL(path_put);
495 #define EMBEDDED_LEVELS 2
500 struct inode *inode; /* path.dentry.d_inode */
502 unsigned seq, m_seq, root_seq;
505 int total_link_count;
512 } *stack, internal[EMBEDDED_LEVELS];
515 static struct nameidata *set_nameidata(struct nameidata *p)
517 struct nameidata *old = current->nameidata;
518 p->stack = p->internal;
519 p->total_link_count = old ? old->total_link_count : 0;
520 current->nameidata = p;
524 static void restore_nameidata(struct nameidata *old)
526 struct nameidata *now = current->nameidata;
528 current->nameidata = old;
530 old->total_link_count = now->total_link_count;
531 if (now->stack != now->internal) {
533 now->stack = now->internal;
537 static int __nd_alloc_stack(struct nameidata *nd)
541 if (nd->flags & LOOKUP_RCU) {
542 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
547 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
552 memcpy(p, nd->internal, sizeof(nd->internal));
557 static inline int nd_alloc_stack(struct nameidata *nd)
559 if (likely(nd->depth != EMBEDDED_LEVELS))
561 if (likely(nd->stack != nd->internal))
563 return __nd_alloc_stack(nd);
566 static void drop_links(struct nameidata *nd)
570 struct saved *last = nd->stack + i;
571 struct inode *inode = last->inode;
572 if (last->cookie && inode->i_op->put_link) {
573 inode->i_op->put_link(inode, last->cookie);
579 static void terminate_walk(struct nameidata *nd)
582 if (!(nd->flags & LOOKUP_RCU)) {
585 for (i = 0; i < nd->depth; i++)
586 path_put(&nd->stack[i].link);
588 nd->flags &= ~LOOKUP_RCU;
589 if (!(nd->flags & LOOKUP_ROOT))
596 /* path_put is needed afterwards regardless of success or failure */
597 static bool legitimize_path(struct nameidata *nd,
598 struct path *path, unsigned seq)
600 int res = __legitimize_mnt(path->mnt, nd->m_seq);
607 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
611 return !read_seqcount_retry(&path->dentry->d_seq, seq);
614 static bool legitimize_links(struct nameidata *nd)
617 for (i = 0; i < nd->depth; i++) {
618 struct saved *last = nd->stack + i;
619 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
629 * Path walking has 2 modes, rcu-walk and ref-walk (see
630 * Documentation/filesystems/path-lookup.txt). In situations when we can't
631 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
632 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
633 * mode. Refcounts are grabbed at the last known good point before rcu-walk
634 * got stuck, so ref-walk may continue from there. If this is not successful
635 * (eg. a seqcount has changed), then failure is returned and it's up to caller
636 * to restart the path walk from the beginning in ref-walk mode.
640 * unlazy_walk - try to switch to ref-walk mode.
641 * @nd: nameidata pathwalk data
642 * @dentry: child of nd->path.dentry or NULL
643 * @seq: seq number to check dentry against
644 * Returns: 0 on success, -ECHILD on failure
646 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
647 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
648 * @nd or NULL. Must be called from rcu-walk context.
649 * Nothing should touch nameidata between unlazy_walk() failure and
652 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
654 struct fs_struct *fs = current->fs;
655 struct dentry *parent = nd->path.dentry;
657 BUG_ON(!(nd->flags & LOOKUP_RCU));
659 nd->flags &= ~LOOKUP_RCU;
660 if (unlikely(!legitimize_links(nd)))
662 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
664 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
668 * For a negative lookup, the lookup sequence point is the parents
669 * sequence point, and it only needs to revalidate the parent dentry.
671 * For a positive lookup, we need to move both the parent and the
672 * dentry from the RCU domain to be properly refcounted. And the
673 * sequence number in the dentry validates *both* dentry counters,
674 * since we checked the sequence number of the parent after we got
675 * the child sequence number. So we know the parent must still
676 * be valid if the child sequence number is still valid.
679 if (read_seqcount_retry(&parent->d_seq, nd->seq))
681 BUG_ON(nd->inode != parent->d_inode);
683 if (!lockref_get_not_dead(&dentry->d_lockref))
685 if (read_seqcount_retry(&dentry->d_seq, seq))
690 * Sequence counts matched. Now make sure that the root is
691 * still valid and get it if required.
693 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
694 spin_lock(&fs->lock);
695 if (unlikely(!path_equal(&nd->root, &fs->root))) {
696 spin_unlock(&fs->lock);
700 spin_unlock(&fs->lock);
713 nd->path.dentry = NULL;
717 if (!(nd->flags & LOOKUP_ROOT))
722 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
724 if (unlikely(!legitimize_path(nd, link, seq))) {
727 nd->flags &= ~LOOKUP_RCU;
729 nd->path.dentry = NULL;
730 if (!(nd->flags & LOOKUP_ROOT))
733 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
740 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
742 return dentry->d_op->d_revalidate(dentry, flags);
746 * complete_walk - successful completion of path walk
747 * @nd: pointer nameidata
749 * If we had been in RCU mode, drop out of it and legitimize nd->path.
750 * Revalidate the final result, unless we'd already done that during
751 * the path walk or the filesystem doesn't ask for it. Return 0 on
752 * success, -error on failure. In case of failure caller does not
753 * need to drop nd->path.
755 static int complete_walk(struct nameidata *nd)
757 struct dentry *dentry = nd->path.dentry;
760 if (nd->flags & LOOKUP_RCU) {
761 if (!(nd->flags & LOOKUP_ROOT))
763 if (unlikely(unlazy_walk(nd, NULL, 0)))
767 if (likely(!(nd->flags & LOOKUP_JUMPED)))
770 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
773 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
783 static __always_inline void set_root(struct nameidata *nd)
785 get_fs_root(current->fs, &nd->root);
788 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
790 struct fs_struct *fs = current->fs;
794 seq = read_seqcount_begin(&fs->seq);
796 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
797 } while (read_seqcount_retry(&fs->seq, seq));
801 static void path_put_conditional(struct path *path, struct nameidata *nd)
804 if (path->mnt != nd->path.mnt)
808 static inline void path_to_nameidata(const struct path *path,
809 struct nameidata *nd)
811 if (!(nd->flags & LOOKUP_RCU)) {
812 dput(nd->path.dentry);
813 if (nd->path.mnt != path->mnt)
814 mntput(nd->path.mnt);
816 nd->path.mnt = path->mnt;
817 nd->path.dentry = path->dentry;
821 * Helper to directly jump to a known parsed path from ->follow_link,
822 * caller must have taken a reference to path beforehand.
824 void nd_jump_link(struct path *path)
826 struct nameidata *nd = current->nameidata;
830 nd->inode = nd->path.dentry->d_inode;
831 nd->flags |= LOOKUP_JUMPED;
834 static inline void put_link(struct nameidata *nd)
836 struct saved *last = nd->stack + --nd->depth;
837 struct inode *inode = last->inode;
838 if (last->cookie && inode->i_op->put_link)
839 inode->i_op->put_link(inode, last->cookie);
840 if (!(nd->flags & LOOKUP_RCU))
841 path_put(&last->link);
844 int sysctl_protected_symlinks __read_mostly = 0;
845 int sysctl_protected_hardlinks __read_mostly = 0;
848 * may_follow_link - Check symlink following for unsafe situations
849 * @nd: nameidata pathwalk data
851 * In the case of the sysctl_protected_symlinks sysctl being enabled,
852 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
853 * in a sticky world-writable directory. This is to protect privileged
854 * processes from failing races against path names that may change out
855 * from under them by way of other users creating malicious symlinks.
856 * It will permit symlinks to be followed only when outside a sticky
857 * world-writable directory, or when the uid of the symlink and follower
858 * match, or when the directory owner matches the symlink's owner.
860 * Returns 0 if following the symlink is allowed, -ve on error.
862 static inline int may_follow_link(struct nameidata *nd)
864 const struct inode *inode;
865 const struct inode *parent;
867 if (!sysctl_protected_symlinks)
870 /* Allowed if owner and follower match. */
871 inode = nd->stack[0].inode;
872 if (uid_eq(current_cred()->fsuid, inode->i_uid))
875 /* Allowed if parent directory not sticky and world-writable. */
876 parent = nd->path.dentry->d_inode;
877 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
880 /* Allowed if parent directory and link owner match. */
881 if (uid_eq(parent->i_uid, inode->i_uid))
884 if (nd->flags & LOOKUP_RCU)
887 audit_log_link_denied("follow_link", &nd->stack[0].link);
892 * safe_hardlink_source - Check for safe hardlink conditions
893 * @inode: the source inode to hardlink from
895 * Return false if at least one of the following conditions:
896 * - inode is not a regular file
898 * - inode is setgid and group-exec
899 * - access failure for read and write
901 * Otherwise returns true.
903 static bool safe_hardlink_source(struct inode *inode)
905 umode_t mode = inode->i_mode;
907 /* Special files should not get pinned to the filesystem. */
911 /* Setuid files should not get pinned to the filesystem. */
915 /* Executable setgid files should not get pinned to the filesystem. */
916 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
919 /* Hardlinking to unreadable or unwritable sources is dangerous. */
920 if (inode_permission(inode, MAY_READ | MAY_WRITE))
927 * may_linkat - Check permissions for creating a hardlink
928 * @link: the source to hardlink from
930 * Block hardlink when all of:
931 * - sysctl_protected_hardlinks enabled
932 * - fsuid does not match inode
933 * - hardlink source is unsafe (see safe_hardlink_source() above)
936 * Returns 0 if successful, -ve on error.
938 static int may_linkat(struct path *link)
940 const struct cred *cred;
943 if (!sysctl_protected_hardlinks)
946 cred = current_cred();
947 inode = link->dentry->d_inode;
949 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
950 * otherwise, it must be a safe source.
952 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
956 audit_log_link_denied("linkat", link);
960 static __always_inline
961 const char *get_link(struct nameidata *nd)
963 struct saved *last = nd->stack + nd->depth - 1;
964 struct dentry *dentry = last->link.dentry;
965 struct inode *inode = last->inode;
969 if (!(nd->flags & LOOKUP_RCU)) {
970 touch_atime(&last->link);
972 } else if (atime_needs_update(&last->link, inode)) {
973 if (unlikely(unlazy_walk(nd, NULL, 0)))
974 return ERR_PTR(-ECHILD);
975 touch_atime(&last->link);
978 error = security_inode_follow_link(dentry, inode,
979 nd->flags & LOOKUP_RCU);
981 return ERR_PTR(error);
983 nd->last_type = LAST_BIND;
986 if (nd->flags & LOOKUP_RCU) {
987 if (unlikely(unlazy_walk(nd, NULL, 0)))
988 return ERR_PTR(-ECHILD);
990 res = inode->i_op->follow_link(dentry, &last->cookie);
991 if (IS_ERR_OR_NULL(res)) {
997 if (nd->flags & LOOKUP_RCU) {
1001 nd->path = nd->root;
1002 d = nd->path.dentry;
1003 nd->inode = d->d_inode;
1004 nd->seq = nd->root_seq;
1005 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
1006 return ERR_PTR(-ECHILD);
1010 path_put(&nd->path);
1011 nd->path = nd->root;
1012 path_get(&nd->root);
1013 nd->inode = nd->path.dentry->d_inode;
1015 nd->flags |= LOOKUP_JUMPED;
1016 while (unlikely(*++res == '/'))
1024 static int follow_up_rcu(struct path *path)
1026 struct mount *mnt = real_mount(path->mnt);
1027 struct mount *parent;
1028 struct dentry *mountpoint;
1030 parent = mnt->mnt_parent;
1031 if (&parent->mnt == path->mnt)
1033 mountpoint = mnt->mnt_mountpoint;
1034 path->dentry = mountpoint;
1035 path->mnt = &parent->mnt;
1040 * follow_up - Find the mountpoint of path's vfsmount
1042 * Given a path, find the mountpoint of its source file system.
1043 * Replace @path with the path of the mountpoint in the parent mount.
1046 * Return 1 if we went up a level and 0 if we were already at the
1049 int follow_up(struct path *path)
1051 struct mount *mnt = real_mount(path->mnt);
1052 struct mount *parent;
1053 struct dentry *mountpoint;
1055 read_seqlock_excl(&mount_lock);
1056 parent = mnt->mnt_parent;
1057 if (parent == mnt) {
1058 read_sequnlock_excl(&mount_lock);
1061 mntget(&parent->mnt);
1062 mountpoint = dget(mnt->mnt_mountpoint);
1063 read_sequnlock_excl(&mount_lock);
1065 path->dentry = mountpoint;
1067 path->mnt = &parent->mnt;
1070 EXPORT_SYMBOL(follow_up);
1073 * Perform an automount
1074 * - return -EISDIR to tell follow_managed() to stop and return the path we
1077 static int follow_automount(struct path *path, struct nameidata *nd,
1080 struct vfsmount *mnt;
1083 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1086 /* We don't want to mount if someone's just doing a stat -
1087 * unless they're stat'ing a directory and appended a '/' to
1090 * We do, however, want to mount if someone wants to open or
1091 * create a file of any type under the mountpoint, wants to
1092 * traverse through the mountpoint or wants to open the
1093 * mounted directory. Also, autofs may mark negative dentries
1094 * as being automount points. These will need the attentions
1095 * of the daemon to instantiate them before they can be used.
1097 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1098 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1099 path->dentry->d_inode)
1102 nd->total_link_count++;
1103 if (nd->total_link_count >= 40)
1106 mnt = path->dentry->d_op->d_automount(path);
1109 * The filesystem is allowed to return -EISDIR here to indicate
1110 * it doesn't want to automount. For instance, autofs would do
1111 * this so that its userspace daemon can mount on this dentry.
1113 * However, we can only permit this if it's a terminal point in
1114 * the path being looked up; if it wasn't then the remainder of
1115 * the path is inaccessible and we should say so.
1117 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1119 return PTR_ERR(mnt);
1122 if (!mnt) /* mount collision */
1125 if (!*need_mntput) {
1126 /* lock_mount() may release path->mnt on error */
1128 *need_mntput = true;
1130 err = finish_automount(mnt, path);
1134 /* Someone else made a mount here whilst we were busy */
1139 path->dentry = dget(mnt->mnt_root);
1148 * Handle a dentry that is managed in some way.
1149 * - Flagged for transit management (autofs)
1150 * - Flagged as mountpoint
1151 * - Flagged as automount point
1153 * This may only be called in refwalk mode.
1155 * Serialization is taken care of in namespace.c
1157 static int follow_managed(struct path *path, struct nameidata *nd)
1159 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1161 bool need_mntput = false;
1164 /* Given that we're not holding a lock here, we retain the value in a
1165 * local variable for each dentry as we look at it so that we don't see
1166 * the components of that value change under us */
1167 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1168 managed &= DCACHE_MANAGED_DENTRY,
1169 unlikely(managed != 0)) {
1170 /* Allow the filesystem to manage the transit without i_mutex
1172 if (managed & DCACHE_MANAGE_TRANSIT) {
1173 BUG_ON(!path->dentry->d_op);
1174 BUG_ON(!path->dentry->d_op->d_manage);
1175 ret = path->dentry->d_op->d_manage(path->dentry, false);
1180 /* Transit to a mounted filesystem. */
1181 if (managed & DCACHE_MOUNTED) {
1182 struct vfsmount *mounted = lookup_mnt(path);
1187 path->mnt = mounted;
1188 path->dentry = dget(mounted->mnt_root);
1193 /* Something is mounted on this dentry in another
1194 * namespace and/or whatever was mounted there in this
1195 * namespace got unmounted before lookup_mnt() could
1199 /* Handle an automount point */
1200 if (managed & DCACHE_NEED_AUTOMOUNT) {
1201 ret = follow_automount(path, nd, &need_mntput);
1207 /* We didn't change the current path point */
1211 if (need_mntput && path->mnt == mnt)
1216 nd->flags |= LOOKUP_JUMPED;
1217 if (unlikely(ret < 0))
1218 path_put_conditional(path, nd);
1222 int follow_down_one(struct path *path)
1224 struct vfsmount *mounted;
1226 mounted = lookup_mnt(path);
1230 path->mnt = mounted;
1231 path->dentry = dget(mounted->mnt_root);
1236 EXPORT_SYMBOL(follow_down_one);
1238 static inline int managed_dentry_rcu(struct dentry *dentry)
1240 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1241 dentry->d_op->d_manage(dentry, true) : 0;
1245 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1246 * we meet a managed dentry that would need blocking.
1248 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1249 struct inode **inode, unsigned *seqp)
1252 struct mount *mounted;
1254 * Don't forget we might have a non-mountpoint managed dentry
1255 * that wants to block transit.
1257 switch (managed_dentry_rcu(path->dentry)) {
1267 if (!d_mountpoint(path->dentry))
1268 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1270 mounted = __lookup_mnt(path->mnt, path->dentry);
1273 path->mnt = &mounted->mnt;
1274 path->dentry = mounted->mnt.mnt_root;
1275 nd->flags |= LOOKUP_JUMPED;
1276 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1278 * Update the inode too. We don't need to re-check the
1279 * dentry sequence number here after this d_inode read,
1280 * because a mount-point is always pinned.
1282 *inode = path->dentry->d_inode;
1284 return !read_seqretry(&mount_lock, nd->m_seq) &&
1285 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1288 static int follow_dotdot_rcu(struct nameidata *nd)
1290 struct inode *inode = nd->inode;
1295 if (nd->path.dentry == nd->root.dentry &&
1296 nd->path.mnt == nd->root.mnt) {
1299 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1300 struct dentry *old = nd->path.dentry;
1301 struct dentry *parent = old->d_parent;
1304 inode = parent->d_inode;
1305 seq = read_seqcount_begin(&parent->d_seq);
1306 if (read_seqcount_retry(&old->d_seq, nd->seq))
1308 nd->path.dentry = parent;
1312 if (!follow_up_rcu(&nd->path))
1314 inode = nd->path.dentry->d_inode;
1315 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1317 while (d_mountpoint(nd->path.dentry)) {
1318 struct mount *mounted;
1319 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1322 nd->path.mnt = &mounted->mnt;
1323 nd->path.dentry = mounted->mnt.mnt_root;
1324 inode = nd->path.dentry->d_inode;
1325 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1326 if (read_seqretry(&mount_lock, nd->m_seq))
1337 * Follow down to the covering mount currently visible to userspace. At each
1338 * point, the filesystem owning that dentry may be queried as to whether the
1339 * caller is permitted to proceed or not.
1341 int follow_down(struct path *path)
1346 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1347 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1348 /* Allow the filesystem to manage the transit without i_mutex
1351 * We indicate to the filesystem if someone is trying to mount
1352 * something here. This gives autofs the chance to deny anyone
1353 * other than its daemon the right to mount on its
1356 * The filesystem may sleep at this point.
1358 if (managed & DCACHE_MANAGE_TRANSIT) {
1359 BUG_ON(!path->dentry->d_op);
1360 BUG_ON(!path->dentry->d_op->d_manage);
1361 ret = path->dentry->d_op->d_manage(
1362 path->dentry, false);
1364 return ret == -EISDIR ? 0 : ret;
1367 /* Transit to a mounted filesystem. */
1368 if (managed & DCACHE_MOUNTED) {
1369 struct vfsmount *mounted = lookup_mnt(path);
1374 path->mnt = mounted;
1375 path->dentry = dget(mounted->mnt_root);
1379 /* Don't handle automount points here */
1384 EXPORT_SYMBOL(follow_down);
1387 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1389 static void follow_mount(struct path *path)
1391 while (d_mountpoint(path->dentry)) {
1392 struct vfsmount *mounted = lookup_mnt(path);
1397 path->mnt = mounted;
1398 path->dentry = dget(mounted->mnt_root);
1402 static void follow_dotdot(struct nameidata *nd)
1408 struct dentry *old = nd->path.dentry;
1410 if (nd->path.dentry == nd->root.dentry &&
1411 nd->path.mnt == nd->root.mnt) {
1414 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1415 /* rare case of legitimate dget_parent()... */
1416 nd->path.dentry = dget_parent(nd->path.dentry);
1420 if (!follow_up(&nd->path))
1423 follow_mount(&nd->path);
1424 nd->inode = nd->path.dentry->d_inode;
1428 * This looks up the name in dcache, possibly revalidates the old dentry and
1429 * allocates a new one if not found or not valid. In the need_lookup argument
1430 * returns whether i_op->lookup is necessary.
1432 * dir->d_inode->i_mutex must be held
1434 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1435 unsigned int flags, bool *need_lookup)
1437 struct dentry *dentry;
1440 *need_lookup = false;
1441 dentry = d_lookup(dir, name);
1443 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1444 error = d_revalidate(dentry, flags);
1445 if (unlikely(error <= 0)) {
1448 return ERR_PTR(error);
1450 d_invalidate(dentry);
1459 dentry = d_alloc(dir, name);
1460 if (unlikely(!dentry))
1461 return ERR_PTR(-ENOMEM);
1463 *need_lookup = true;
1469 * Call i_op->lookup on the dentry. The dentry must be negative and
1472 * dir->d_inode->i_mutex must be held
1474 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1479 /* Don't create child dentry for a dead directory. */
1480 if (unlikely(IS_DEADDIR(dir))) {
1482 return ERR_PTR(-ENOENT);
1485 old = dir->i_op->lookup(dir, dentry, flags);
1486 if (unlikely(old)) {
1493 static struct dentry *__lookup_hash(struct qstr *name,
1494 struct dentry *base, unsigned int flags)
1497 struct dentry *dentry;
1499 dentry = lookup_dcache(name, base, flags, &need_lookup);
1503 return lookup_real(base->d_inode, dentry, flags);
1507 * It's more convoluted than I'd like it to be, but... it's still fairly
1508 * small and for now I'd prefer to have fast path as straight as possible.
1509 * It _is_ time-critical.
1511 static int lookup_fast(struct nameidata *nd,
1512 struct path *path, struct inode **inode,
1515 struct vfsmount *mnt = nd->path.mnt;
1516 struct dentry *dentry, *parent = nd->path.dentry;
1522 * Rename seqlock is not required here because in the off chance
1523 * of a false negative due to a concurrent rename, we're going to
1524 * do the non-racy lookup, below.
1526 if (nd->flags & LOOKUP_RCU) {
1529 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1534 * This sequence count validates that the inode matches
1535 * the dentry name information from lookup.
1537 *inode = d_backing_inode(dentry);
1538 negative = d_is_negative(dentry);
1539 if (read_seqcount_retry(&dentry->d_seq, seq))
1545 * This sequence count validates that the parent had no
1546 * changes while we did the lookup of the dentry above.
1548 * The memory barrier in read_seqcount_begin of child is
1549 * enough, we can use __read_seqcount_retry here.
1551 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1555 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1556 status = d_revalidate(dentry, nd->flags);
1557 if (unlikely(status <= 0)) {
1558 if (status != -ECHILD)
1564 path->dentry = dentry;
1565 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1568 if (unlazy_walk(nd, dentry, seq))
1571 dentry = __d_lookup(parent, &nd->last);
1574 if (unlikely(!dentry))
1577 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1578 status = d_revalidate(dentry, nd->flags);
1579 if (unlikely(status <= 0)) {
1584 d_invalidate(dentry);
1589 if (unlikely(d_is_negative(dentry))) {
1594 path->dentry = dentry;
1595 err = follow_managed(path, nd);
1597 *inode = d_backing_inode(path->dentry);
1604 /* Fast lookup failed, do it the slow way */
1605 static int lookup_slow(struct nameidata *nd, struct path *path)
1607 struct dentry *dentry, *parent;
1609 parent = nd->path.dentry;
1610 BUG_ON(nd->inode != parent->d_inode);
1612 mutex_lock(&parent->d_inode->i_mutex);
1613 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1614 mutex_unlock(&parent->d_inode->i_mutex);
1616 return PTR_ERR(dentry);
1617 path->mnt = nd->path.mnt;
1618 path->dentry = dentry;
1619 return follow_managed(path, nd);
1622 static inline int may_lookup(struct nameidata *nd)
1624 if (nd->flags & LOOKUP_RCU) {
1625 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1628 if (unlazy_walk(nd, NULL, 0))
1631 return inode_permission(nd->inode, MAY_EXEC);
1634 static inline int handle_dots(struct nameidata *nd, int type)
1636 if (type == LAST_DOTDOT) {
1637 if (nd->flags & LOOKUP_RCU) {
1638 return follow_dotdot_rcu(nd);
1645 static int pick_link(struct nameidata *nd, struct path *link,
1646 struct inode *inode, unsigned seq)
1650 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1651 path_to_nameidata(link, nd);
1654 if (!(nd->flags & LOOKUP_RCU)) {
1655 if (link->mnt == nd->path.mnt)
1658 error = nd_alloc_stack(nd);
1659 if (unlikely(error)) {
1660 if (error == -ECHILD) {
1661 if (unlikely(unlazy_link(nd, link, seq)))
1663 error = nd_alloc_stack(nd);
1671 last = nd->stack + nd->depth++;
1673 last->cookie = NULL;
1674 last->inode = inode;
1680 * Do we need to follow links? We _really_ want to be able
1681 * to do this check without having to look at inode->i_op,
1682 * so we keep a cache of "no, this doesn't need follow_link"
1683 * for the common case.
1685 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1687 struct inode *inode, unsigned seq)
1689 if (likely(!d_is_symlink(link->dentry)))
1693 return pick_link(nd, link, inode, seq);
1696 enum {WALK_GET = 1, WALK_PUT = 2};
1698 static int walk_component(struct nameidata *nd, int flags)
1701 struct inode *inode;
1705 * "." and ".." are special - ".." especially so because it has
1706 * to be able to know about the current root directory and
1707 * parent relationships.
1709 if (unlikely(nd->last_type != LAST_NORM)) {
1710 err = handle_dots(nd, nd->last_type);
1711 if (flags & WALK_PUT)
1715 err = lookup_fast(nd, &path, &inode, &seq);
1716 if (unlikely(err)) {
1720 err = lookup_slow(nd, &path);
1724 inode = d_backing_inode(path.dentry);
1725 seq = 0; /* we are already out of RCU mode */
1727 if (d_is_negative(path.dentry))
1731 if (flags & WALK_PUT)
1733 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1736 path_to_nameidata(&path, nd);
1742 path_to_nameidata(&path, nd);
1747 * We can do the critical dentry name comparison and hashing
1748 * operations one word at a time, but we are limited to:
1750 * - Architectures with fast unaligned word accesses. We could
1751 * do a "get_unaligned()" if this helps and is sufficiently
1754 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1755 * do not trap on the (extremely unlikely) case of a page
1756 * crossing operation.
1758 * - Furthermore, we need an efficient 64-bit compile for the
1759 * 64-bit case in order to generate the "number of bytes in
1760 * the final mask". Again, that could be replaced with a
1761 * efficient population count instruction or similar.
1763 #ifdef CONFIG_DCACHE_WORD_ACCESS
1765 #include <asm/word-at-a-time.h>
1769 static inline unsigned int fold_hash(unsigned long hash)
1771 return hash_64(hash, 32);
1774 #else /* 32-bit case */
1776 #define fold_hash(x) (x)
1780 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1782 unsigned long a, mask;
1783 unsigned long hash = 0;
1786 a = load_unaligned_zeropad(name);
1787 if (len < sizeof(unsigned long))
1791 name += sizeof(unsigned long);
1792 len -= sizeof(unsigned long);
1796 mask = bytemask_from_count(len);
1799 return fold_hash(hash);
1801 EXPORT_SYMBOL(full_name_hash);
1804 * Calculate the length and hash of the path component, and
1805 * return the "hash_len" as the result.
1807 static inline u64 hash_name(const char *name)
1809 unsigned long a, b, adata, bdata, mask, hash, len;
1810 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1813 len = -sizeof(unsigned long);
1815 hash = (hash + a) * 9;
1816 len += sizeof(unsigned long);
1817 a = load_unaligned_zeropad(name+len);
1818 b = a ^ REPEAT_BYTE('/');
1819 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1821 adata = prep_zero_mask(a, adata, &constants);
1822 bdata = prep_zero_mask(b, bdata, &constants);
1824 mask = create_zero_mask(adata | bdata);
1826 hash += a & zero_bytemask(mask);
1827 len += find_zero(mask);
1828 return hashlen_create(fold_hash(hash), len);
1833 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1835 unsigned long hash = init_name_hash();
1837 hash = partial_name_hash(*name++, hash);
1838 return end_name_hash(hash);
1840 EXPORT_SYMBOL(full_name_hash);
1843 * We know there's a real path component here of at least
1846 static inline u64 hash_name(const char *name)
1848 unsigned long hash = init_name_hash();
1849 unsigned long len = 0, c;
1851 c = (unsigned char)*name;
1854 hash = partial_name_hash(c, hash);
1855 c = (unsigned char)name[len];
1856 } while (c && c != '/');
1857 return hashlen_create(end_name_hash(hash), len);
1864 * This is the basic name resolution function, turning a pathname into
1865 * the final dentry. We expect 'base' to be positive and a directory.
1867 * Returns 0 and nd will have valid dentry and mnt on success.
1868 * Returns error and drops reference to input namei data on failure.
1870 static int link_path_walk(const char *name, struct nameidata *nd)
1879 /* At this point we know we have a real path component. */
1884 err = may_lookup(nd);
1888 hash_len = hash_name(name);
1891 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1893 if (name[1] == '.') {
1895 nd->flags |= LOOKUP_JUMPED;
1901 if (likely(type == LAST_NORM)) {
1902 struct dentry *parent = nd->path.dentry;
1903 nd->flags &= ~LOOKUP_JUMPED;
1904 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1905 struct qstr this = { { .hash_len = hash_len }, .name = name };
1906 err = parent->d_op->d_hash(parent, &this);
1909 hash_len = this.hash_len;
1914 nd->last.hash_len = hash_len;
1915 nd->last.name = name;
1916 nd->last_type = type;
1918 name += hashlen_len(hash_len);
1922 * If it wasn't NUL, we know it was '/'. Skip that
1923 * slash, and continue until no more slashes.
1927 } while (unlikely(*name == '/'));
1928 if (unlikely(!*name)) {
1930 /* pathname body, done */
1933 name = nd->stack[nd->depth - 1].name;
1934 /* trailing symlink, done */
1937 /* last component of nested symlink */
1938 err = walk_component(nd, WALK_GET | WALK_PUT);
1940 err = walk_component(nd, WALK_GET);
1946 const char *s = get_link(nd);
1948 if (unlikely(IS_ERR(s)))
1955 nd->stack[nd->depth - 1].name = name;
1960 if (unlikely(!d_can_lookup(nd->path.dentry)))
1965 static const char *path_init(int dfd, const struct filename *name,
1966 unsigned int flags, struct nameidata *nd)
1969 const char *s = name->name;
1971 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1972 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1974 nd->total_link_count = 0;
1975 if (flags & LOOKUP_ROOT) {
1976 struct dentry *root = nd->root.dentry;
1977 struct inode *inode = root->d_inode;
1979 if (!d_can_lookup(root))
1980 return ERR_PTR(-ENOTDIR);
1981 retval = inode_permission(inode, MAY_EXEC);
1983 return ERR_PTR(retval);
1985 nd->path = nd->root;
1987 if (flags & LOOKUP_RCU) {
1989 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1990 nd->root_seq = nd->seq;
1991 nd->m_seq = read_seqbegin(&mount_lock);
1993 path_get(&nd->path);
1998 nd->root.mnt = NULL;
2000 nd->m_seq = read_seqbegin(&mount_lock);
2002 if (flags & LOOKUP_RCU) {
2004 nd->seq = set_root_rcu(nd);
2007 path_get(&nd->root);
2009 nd->path = nd->root;
2010 } else if (dfd == AT_FDCWD) {
2011 if (flags & LOOKUP_RCU) {
2012 struct fs_struct *fs = current->fs;
2018 seq = read_seqcount_begin(&fs->seq);
2020 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2021 } while (read_seqcount_retry(&fs->seq, seq));
2023 get_fs_pwd(current->fs, &nd->path);
2026 /* Caller must check execute permissions on the starting path component */
2027 struct fd f = fdget_raw(dfd);
2028 struct dentry *dentry;
2031 return ERR_PTR(-EBADF);
2033 dentry = f.file->f_path.dentry;
2036 if (!d_can_lookup(dentry)) {
2038 return ERR_PTR(-ENOTDIR);
2042 nd->path = f.file->f_path;
2043 if (flags & LOOKUP_RCU) {
2045 nd->inode = nd->path.dentry->d_inode;
2046 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2048 path_get(&nd->path);
2049 nd->inode = nd->path.dentry->d_inode;
2055 nd->inode = nd->path.dentry->d_inode;
2056 if (!(flags & LOOKUP_RCU))
2058 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
2060 if (!(nd->flags & LOOKUP_ROOT))
2061 nd->root.mnt = NULL;
2063 return ERR_PTR(-ECHILD);
2066 static void path_cleanup(struct nameidata *nd)
2068 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2069 path_put(&nd->root);
2070 nd->root.mnt = NULL;
2074 static const char *trailing_symlink(struct nameidata *nd)
2077 int error = may_follow_link(nd);
2078 if (unlikely(error))
2079 return ERR_PTR(error);
2080 nd->flags |= LOOKUP_PARENT;
2081 nd->stack[0].name = NULL;
2086 static inline int lookup_last(struct nameidata *nd)
2088 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2089 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2091 nd->flags &= ~LOOKUP_PARENT;
2092 return walk_component(nd,
2093 nd->flags & LOOKUP_FOLLOW
2095 ? WALK_PUT | WALK_GET
2100 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2101 static int path_lookupat(int dfd, const struct filename *name,
2102 unsigned int flags, struct nameidata *nd)
2104 const char *s = path_init(dfd, name, flags, nd);
2109 while (!(err = link_path_walk(s, nd))
2110 && ((err = lookup_last(nd)) > 0)) {
2111 s = trailing_symlink(nd);
2118 err = complete_walk(nd);
2120 if (!err && nd->flags & LOOKUP_DIRECTORY)
2121 if (!d_can_lookup(nd->path.dentry))
2130 static int filename_lookup(int dfd, struct filename *name,
2131 unsigned int flags, struct nameidata *nd)
2134 struct nameidata *saved_nd = set_nameidata(nd);
2136 retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2137 if (unlikely(retval == -ECHILD))
2138 retval = path_lookupat(dfd, name, flags, nd);
2139 if (unlikely(retval == -ESTALE))
2140 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2142 if (likely(!retval))
2143 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2144 restore_nameidata(saved_nd);
2148 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2149 static int path_parentat(int dfd, const struct filename *name,
2150 unsigned int flags, struct nameidata *nd)
2152 const char *s = path_init(dfd, name, flags, nd);
2156 err = link_path_walk(s, nd);
2158 err = complete_walk(nd);
2165 static int filename_parentat(int dfd, struct filename *name,
2166 unsigned int flags, struct nameidata *nd)
2169 struct nameidata *saved_nd = set_nameidata(nd);
2171 retval = path_parentat(dfd, name, flags | LOOKUP_RCU, nd);
2172 if (unlikely(retval == -ECHILD))
2173 retval = path_parentat(dfd, name, flags, nd);
2174 if (unlikely(retval == -ESTALE))
2175 retval = path_parentat(dfd, name, flags | LOOKUP_REVAL, nd);
2177 if (likely(!retval))
2178 audit_inode(name, nd->path.dentry, LOOKUP_PARENT);
2179 restore_nameidata(saved_nd);
2183 /* does lookup, returns the object with parent locked */
2184 struct dentry *kern_path_locked(const char *name, struct path *path)
2186 struct filename *filename = getname_kernel(name);
2187 struct nameidata nd;
2191 if (IS_ERR(filename))
2192 return ERR_CAST(filename);
2194 err = filename_parentat(AT_FDCWD, filename, 0, &nd);
2199 if (nd.last_type != LAST_NORM) {
2201 d = ERR_PTR(-EINVAL);
2204 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2205 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2207 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2217 int kern_path(const char *name, unsigned int flags, struct path *path)
2219 struct nameidata nd;
2220 struct filename *filename = getname_kernel(name);
2221 int res = PTR_ERR(filename);
2223 if (!IS_ERR(filename)) {
2224 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2231 EXPORT_SYMBOL(kern_path);
2234 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2235 * @dentry: pointer to dentry of the base directory
2236 * @mnt: pointer to vfs mount of the base directory
2237 * @name: pointer to file name
2238 * @flags: lookup flags
2239 * @path: pointer to struct path to fill
2241 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2242 const char *name, unsigned int flags,
2245 struct filename *filename = getname_kernel(name);
2246 int err = PTR_ERR(filename);
2248 BUG_ON(flags & LOOKUP_PARENT);
2250 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2251 if (!IS_ERR(filename)) {
2252 struct nameidata nd;
2253 nd.root.dentry = dentry;
2255 err = filename_lookup(AT_FDCWD, filename,
2256 flags | LOOKUP_ROOT, &nd);
2263 EXPORT_SYMBOL(vfs_path_lookup);
2266 * lookup_one_len - filesystem helper to lookup single pathname component
2267 * @name: pathname component to lookup
2268 * @base: base directory to lookup from
2269 * @len: maximum length @len should be interpreted to
2271 * Note that this routine is purely a helper for filesystem usage and should
2272 * not be called by generic code.
2274 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2280 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2284 this.hash = full_name_hash(name, len);
2286 return ERR_PTR(-EACCES);
2288 if (unlikely(name[0] == '.')) {
2289 if (len < 2 || (len == 2 && name[1] == '.'))
2290 return ERR_PTR(-EACCES);
2294 c = *(const unsigned char *)name++;
2295 if (c == '/' || c == '\0')
2296 return ERR_PTR(-EACCES);
2299 * See if the low-level filesystem might want
2300 * to use its own hash..
2302 if (base->d_flags & DCACHE_OP_HASH) {
2303 int err = base->d_op->d_hash(base, &this);
2305 return ERR_PTR(err);
2308 err = inode_permission(base->d_inode, MAY_EXEC);
2310 return ERR_PTR(err);
2312 return __lookup_hash(&this, base, 0);
2314 EXPORT_SYMBOL(lookup_one_len);
2316 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2317 struct path *path, int *empty)
2319 struct nameidata nd;
2320 struct filename *tmp = getname_flags(name, flags, empty);
2321 int err = PTR_ERR(tmp);
2324 BUG_ON(flags & LOOKUP_PARENT);
2326 err = filename_lookup(dfd, tmp, flags, &nd);
2334 int user_path_at(int dfd, const char __user *name, unsigned flags,
2337 return user_path_at_empty(dfd, name, flags, path, NULL);
2339 EXPORT_SYMBOL(user_path_at);
2342 * NB: most callers don't do anything directly with the reference to the
2343 * to struct filename, but the nd->last pointer points into the name string
2344 * allocated by getname. So we must hold the reference to it until all
2345 * path-walking is complete.
2347 static struct filename *
2348 user_path_parent(int dfd, const char __user *path,
2349 struct path *parent,
2354 struct nameidata nd;
2355 struct filename *s = getname(path);
2358 /* only LOOKUP_REVAL is allowed in extra flags */
2359 flags &= LOOKUP_REVAL;
2364 error = filename_parentat(dfd, s, flags, &nd);
2367 return ERR_PTR(error);
2371 *type = nd.last_type;
2377 * mountpoint_last - look up last component for umount
2378 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2379 * @path: pointer to container for result
2381 * This is a special lookup_last function just for umount. In this case, we
2382 * need to resolve the path without doing any revalidation.
2384 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2385 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2386 * in almost all cases, this lookup will be served out of the dcache. The only
2387 * cases where it won't are if nd->last refers to a symlink or the path is
2388 * bogus and it doesn't exist.
2391 * -error: if there was an error during lookup. This includes -ENOENT if the
2392 * lookup found a negative dentry. The nd->path reference will also be
2395 * 0: if we successfully resolved nd->path and found it to not to be a
2396 * symlink that needs to be followed. "path" will also be populated.
2397 * The nd->path reference will also be put.
2399 * 1: if we successfully resolved nd->last and found it to be a symlink
2400 * that needs to be followed. "path" will be populated with the path
2401 * to the link, and nd->path will *not* be put.
2404 mountpoint_last(struct nameidata *nd, struct path *path)
2407 struct dentry *dentry;
2408 struct dentry *dir = nd->path.dentry;
2410 /* If we're in rcuwalk, drop out of it to handle last component */
2411 if (nd->flags & LOOKUP_RCU) {
2412 if (unlazy_walk(nd, NULL, 0))
2416 nd->flags &= ~LOOKUP_PARENT;
2418 if (unlikely(nd->last_type != LAST_NORM)) {
2419 error = handle_dots(nd, nd->last_type);
2422 dentry = dget(nd->path.dentry);
2426 mutex_lock(&dir->d_inode->i_mutex);
2427 dentry = d_lookup(dir, &nd->last);
2430 * No cached dentry. Mounted dentries are pinned in the cache,
2431 * so that means that this dentry is probably a symlink or the
2432 * path doesn't actually point to a mounted dentry.
2434 dentry = d_alloc(dir, &nd->last);
2436 mutex_unlock(&dir->d_inode->i_mutex);
2439 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2440 if (IS_ERR(dentry)) {
2441 mutex_unlock(&dir->d_inode->i_mutex);
2442 return PTR_ERR(dentry);
2445 mutex_unlock(&dir->d_inode->i_mutex);
2448 if (d_is_negative(dentry)) {
2454 path->dentry = dentry;
2455 path->mnt = nd->path.mnt;
2456 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2457 d_backing_inode(dentry), 0);
2458 if (unlikely(error))
2466 * path_mountpoint - look up a path to be umounted
2467 * @dfd: directory file descriptor to start walk from
2468 * @name: full pathname to walk
2469 * @path: pointer to container for result
2470 * @flags: lookup flags
2472 * Look up the given name, but don't attempt to revalidate the last component.
2473 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2476 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2477 struct nameidata *nd, unsigned int flags)
2479 const char *s = path_init(dfd, name, flags, nd);
2483 while (!(err = link_path_walk(s, nd)) &&
2484 (err = mountpoint_last(nd, path)) > 0) {
2485 s = trailing_symlink(nd);
2497 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2500 struct nameidata nd, *saved;
2503 return PTR_ERR(name);
2504 saved = set_nameidata(&nd);
2505 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2506 if (unlikely(error == -ECHILD))
2507 error = path_mountpoint(dfd, name, path, &nd, flags);
2508 if (unlikely(error == -ESTALE))
2509 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2511 audit_inode(name, path->dentry, 0);
2512 restore_nameidata(saved);
2518 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2519 * @dfd: directory file descriptor
2520 * @name: pathname from userland
2521 * @flags: lookup flags
2522 * @path: pointer to container to hold result
2524 * A umount is a special case for path walking. We're not actually interested
2525 * in the inode in this situation, and ESTALE errors can be a problem. We
2526 * simply want track down the dentry and vfsmount attached at the mountpoint
2527 * and avoid revalidating the last component.
2529 * Returns 0 and populates "path" on success.
2532 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2535 return filename_mountpoint(dfd, getname(name), path, flags);
2539 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2542 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2544 EXPORT_SYMBOL(kern_path_mountpoint);
2546 int __check_sticky(struct inode *dir, struct inode *inode)
2548 kuid_t fsuid = current_fsuid();
2550 if (uid_eq(inode->i_uid, fsuid))
2552 if (uid_eq(dir->i_uid, fsuid))
2554 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2556 EXPORT_SYMBOL(__check_sticky);
2559 * Check whether we can remove a link victim from directory dir, check
2560 * whether the type of victim is right.
2561 * 1. We can't do it if dir is read-only (done in permission())
2562 * 2. We should have write and exec permissions on dir
2563 * 3. We can't remove anything from append-only dir
2564 * 4. We can't do anything with immutable dir (done in permission())
2565 * 5. If the sticky bit on dir is set we should either
2566 * a. be owner of dir, or
2567 * b. be owner of victim, or
2568 * c. have CAP_FOWNER capability
2569 * 6. If the victim is append-only or immutable we can't do antyhing with
2570 * links pointing to it.
2571 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2572 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2573 * 9. We can't remove a root or mountpoint.
2574 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2575 * nfs_async_unlink().
2577 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2579 struct inode *inode = d_backing_inode(victim);
2582 if (d_is_negative(victim))
2586 BUG_ON(victim->d_parent->d_inode != dir);
2587 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2589 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2595 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2596 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2599 if (!d_is_dir(victim))
2601 if (IS_ROOT(victim))
2603 } else if (d_is_dir(victim))
2605 if (IS_DEADDIR(dir))
2607 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2612 /* Check whether we can create an object with dentry child in directory
2614 * 1. We can't do it if child already exists (open has special treatment for
2615 * this case, but since we are inlined it's OK)
2616 * 2. We can't do it if dir is read-only (done in permission())
2617 * 3. We should have write and exec permissions on dir
2618 * 4. We can't do it if dir is immutable (done in permission())
2620 static inline int may_create(struct inode *dir, struct dentry *child)
2622 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2625 if (IS_DEADDIR(dir))
2627 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2631 * p1 and p2 should be directories on the same fs.
2633 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2638 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2642 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2644 p = d_ancestor(p2, p1);
2646 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2647 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2651 p = d_ancestor(p1, p2);
2653 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2654 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2658 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2659 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2662 EXPORT_SYMBOL(lock_rename);
2664 void unlock_rename(struct dentry *p1, struct dentry *p2)
2666 mutex_unlock(&p1->d_inode->i_mutex);
2668 mutex_unlock(&p2->d_inode->i_mutex);
2669 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2672 EXPORT_SYMBOL(unlock_rename);
2674 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2677 int error = may_create(dir, dentry);
2681 if (!dir->i_op->create)
2682 return -EACCES; /* shouldn't it be ENOSYS? */
2685 error = security_inode_create(dir, dentry, mode);
2688 error = dir->i_op->create(dir, dentry, mode, want_excl);
2690 fsnotify_create(dir, dentry);
2693 EXPORT_SYMBOL(vfs_create);
2695 static int may_open(struct path *path, int acc_mode, int flag)
2697 struct dentry *dentry = path->dentry;
2698 struct inode *inode = dentry->d_inode;
2708 switch (inode->i_mode & S_IFMT) {
2712 if (acc_mode & MAY_WRITE)
2717 if (path->mnt->mnt_flags & MNT_NODEV)
2726 error = inode_permission(inode, acc_mode);
2731 * An append-only file must be opened in append mode for writing.
2733 if (IS_APPEND(inode)) {
2734 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2740 /* O_NOATIME can only be set by the owner or superuser */
2741 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2747 static int handle_truncate(struct file *filp)
2749 struct path *path = &filp->f_path;
2750 struct inode *inode = path->dentry->d_inode;
2751 int error = get_write_access(inode);
2755 * Refuse to truncate files with mandatory locks held on them.
2757 error = locks_verify_locked(filp);
2759 error = security_path_truncate(path);
2761 error = do_truncate(path->dentry, 0,
2762 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2765 put_write_access(inode);
2769 static inline int open_to_namei_flags(int flag)
2771 if ((flag & O_ACCMODE) == 3)
2776 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2778 int error = security_path_mknod(dir, dentry, mode, 0);
2782 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2786 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2790 * Attempt to atomically look up, create and open a file from a negative
2793 * Returns 0 if successful. The file will have been created and attached to
2794 * @file by the filesystem calling finish_open().
2796 * Returns 1 if the file was looked up only or didn't need creating. The
2797 * caller will need to perform the open themselves. @path will have been
2798 * updated to point to the new dentry. This may be negative.
2800 * Returns an error code otherwise.
2802 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2803 struct path *path, struct file *file,
2804 const struct open_flags *op,
2805 bool got_write, bool need_lookup,
2808 struct inode *dir = nd->path.dentry->d_inode;
2809 unsigned open_flag = open_to_namei_flags(op->open_flag);
2813 int create_error = 0;
2814 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2817 BUG_ON(dentry->d_inode);
2819 /* Don't create child dentry for a dead directory. */
2820 if (unlikely(IS_DEADDIR(dir))) {
2826 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2827 mode &= ~current_umask();
2829 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2831 open_flag &= ~O_TRUNC;
2834 * Checking write permission is tricky, bacuse we don't know if we are
2835 * going to actually need it: O_CREAT opens should work as long as the
2836 * file exists. But checking existence breaks atomicity. The trick is
2837 * to check access and if not granted clear O_CREAT from the flags.
2839 * Another problem is returing the "right" error value (e.g. for an
2840 * O_EXCL open we want to return EEXIST not EROFS).
2842 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2843 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2844 if (!(open_flag & O_CREAT)) {
2846 * No O_CREATE -> atomicity not a requirement -> fall
2847 * back to lookup + open
2850 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2851 /* Fall back and fail with the right error */
2852 create_error = -EROFS;
2855 /* No side effects, safe to clear O_CREAT */
2856 create_error = -EROFS;
2857 open_flag &= ~O_CREAT;
2861 if (open_flag & O_CREAT) {
2862 error = may_o_create(&nd->path, dentry, mode);
2864 create_error = error;
2865 if (open_flag & O_EXCL)
2867 open_flag &= ~O_CREAT;
2871 if (nd->flags & LOOKUP_DIRECTORY)
2872 open_flag |= O_DIRECTORY;
2874 file->f_path.dentry = DENTRY_NOT_SET;
2875 file->f_path.mnt = nd->path.mnt;
2876 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2879 if (create_error && error == -ENOENT)
2880 error = create_error;
2884 if (error) { /* returned 1, that is */
2885 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2889 if (file->f_path.dentry) {
2891 dentry = file->f_path.dentry;
2893 if (*opened & FILE_CREATED)
2894 fsnotify_create(dir, dentry);
2895 if (!dentry->d_inode) {
2896 WARN_ON(*opened & FILE_CREATED);
2898 error = create_error;
2902 if (excl && !(*opened & FILE_CREATED)) {
2911 * We didn't have the inode before the open, so check open permission
2914 acc_mode = op->acc_mode;
2915 if (*opened & FILE_CREATED) {
2916 WARN_ON(!(open_flag & O_CREAT));
2917 fsnotify_create(dir, dentry);
2918 acc_mode = MAY_OPEN;
2920 error = may_open(&file->f_path, acc_mode, open_flag);
2930 dentry = lookup_real(dir, dentry, nd->flags);
2932 return PTR_ERR(dentry);
2935 int open_flag = op->open_flag;
2937 error = create_error;
2938 if ((open_flag & O_EXCL)) {
2939 if (!dentry->d_inode)
2941 } else if (!dentry->d_inode) {
2943 } else if ((open_flag & O_TRUNC) &&
2947 /* will fail later, go on to get the right error */
2951 path->dentry = dentry;
2952 path->mnt = nd->path.mnt;
2957 * Look up and maybe create and open the last component.
2959 * Must be called with i_mutex held on parent.
2961 * Returns 0 if the file was successfully atomically created (if necessary) and
2962 * opened. In this case the file will be returned attached to @file.
2964 * Returns 1 if the file was not completely opened at this time, though lookups
2965 * and creations will have been performed and the dentry returned in @path will
2966 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2967 * specified then a negative dentry may be returned.
2969 * An error code is returned otherwise.
2971 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2972 * cleared otherwise prior to returning.
2974 static int lookup_open(struct nameidata *nd, struct path *path,
2976 const struct open_flags *op,
2977 bool got_write, int *opened)
2979 struct dentry *dir = nd->path.dentry;
2980 struct inode *dir_inode = dir->d_inode;
2981 struct dentry *dentry;
2985 *opened &= ~FILE_CREATED;
2986 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2988 return PTR_ERR(dentry);
2990 /* Cached positive dentry: will open in f_op->open */
2991 if (!need_lookup && dentry->d_inode)
2994 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2995 return atomic_open(nd, dentry, path, file, op, got_write,
2996 need_lookup, opened);
3000 BUG_ON(dentry->d_inode);
3002 dentry = lookup_real(dir_inode, dentry, nd->flags);
3004 return PTR_ERR(dentry);
3007 /* Negative dentry, just create the file */
3008 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
3009 umode_t mode = op->mode;
3010 if (!IS_POSIXACL(dir->d_inode))
3011 mode &= ~current_umask();
3013 * This write is needed to ensure that a
3014 * rw->ro transition does not occur between
3015 * the time when the file is created and when
3016 * a permanent write count is taken through
3017 * the 'struct file' in finish_open().
3023 *opened |= FILE_CREATED;
3024 error = security_path_mknod(&nd->path, dentry, mode, 0);
3027 error = vfs_create(dir->d_inode, dentry, mode,
3028 nd->flags & LOOKUP_EXCL);
3033 path->dentry = dentry;
3034 path->mnt = nd->path.mnt;
3043 * Handle the last step of open()
3045 static int do_last(struct nameidata *nd,
3046 struct file *file, const struct open_flags *op,
3047 int *opened, struct filename *name)
3049 struct dentry *dir = nd->path.dentry;
3050 int open_flag = op->open_flag;
3051 bool will_truncate = (open_flag & O_TRUNC) != 0;
3052 bool got_write = false;
3053 int acc_mode = op->acc_mode;
3055 struct inode *inode;
3056 struct path save_parent = { .dentry = NULL, .mnt = NULL };
3058 bool retried = false;
3061 nd->flags &= ~LOOKUP_PARENT;
3062 nd->flags |= op->intent;
3064 if (nd->last_type != LAST_NORM) {
3065 error = handle_dots(nd, nd->last_type);
3066 if (unlikely(error))
3071 if (!(open_flag & O_CREAT)) {
3072 if (nd->last.name[nd->last.len])
3073 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3074 /* we _can_ be in RCU mode here */
3075 error = lookup_fast(nd, &path, &inode, &seq);
3082 BUG_ON(nd->inode != dir->d_inode);
3084 /* create side of things */
3086 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3087 * has been cleared when we got to the last component we are
3090 error = complete_walk(nd);
3094 audit_inode(name, dir, LOOKUP_PARENT);
3095 /* trailing slashes? */
3096 if (unlikely(nd->last.name[nd->last.len]))
3101 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3102 error = mnt_want_write(nd->path.mnt);
3106 * do _not_ fail yet - we might not need that or fail with
3107 * a different error; let lookup_open() decide; we'll be
3108 * dropping this one anyway.
3111 mutex_lock(&dir->d_inode->i_mutex);
3112 error = lookup_open(nd, &path, file, op, got_write, opened);
3113 mutex_unlock(&dir->d_inode->i_mutex);
3119 if ((*opened & FILE_CREATED) ||
3120 !S_ISREG(file_inode(file)->i_mode))
3121 will_truncate = false;
3123 audit_inode(name, file->f_path.dentry, 0);
3127 if (*opened & FILE_CREATED) {
3128 /* Don't check for write permission, don't truncate */
3129 open_flag &= ~O_TRUNC;
3130 will_truncate = false;
3131 acc_mode = MAY_OPEN;
3132 path_to_nameidata(&path, nd);
3133 goto finish_open_created;
3137 * create/update audit record if it already exists.
3139 if (d_is_positive(path.dentry))
3140 audit_inode(name, path.dentry, 0);
3143 * If atomic_open() acquired write access it is dropped now due to
3144 * possible mount and symlink following (this might be optimized away if
3148 mnt_drop_write(nd->path.mnt);
3152 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3153 path_to_nameidata(&path, nd);
3157 error = follow_managed(&path, nd);
3158 if (unlikely(error < 0))
3161 BUG_ON(nd->flags & LOOKUP_RCU);
3162 inode = d_backing_inode(path.dentry);
3163 seq = 0; /* out of RCU mode, so the value doesn't matter */
3164 if (unlikely(d_is_negative(path.dentry))) {
3165 path_to_nameidata(&path, nd);
3171 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3173 if (unlikely(error))
3176 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3177 path_to_nameidata(&path, nd);
3181 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3182 path_to_nameidata(&path, nd);
3184 save_parent.dentry = nd->path.dentry;
3185 save_parent.mnt = mntget(path.mnt);
3186 nd->path.dentry = path.dentry;
3191 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3193 error = complete_walk(nd);
3195 path_put(&save_parent);
3198 audit_inode(name, nd->path.dentry, 0);
3200 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3203 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3205 if (!d_is_reg(nd->path.dentry))
3206 will_truncate = false;
3208 if (will_truncate) {
3209 error = mnt_want_write(nd->path.mnt);
3214 finish_open_created:
3215 error = may_open(&nd->path, acc_mode, open_flag);
3219 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3220 error = vfs_open(&nd->path, file, current_cred());
3222 *opened |= FILE_OPENED;
3224 if (error == -EOPENSTALE)
3229 error = open_check_o_direct(file);
3232 error = ima_file_check(file, op->acc_mode, *opened);
3236 if (will_truncate) {
3237 error = handle_truncate(file);
3243 mnt_drop_write(nd->path.mnt);
3244 path_put(&save_parent);
3252 /* If no saved parent or already retried then can't retry */
3253 if (!save_parent.dentry || retried)
3256 BUG_ON(save_parent.dentry != dir);
3257 path_put(&nd->path);
3258 nd->path = save_parent;
3259 nd->inode = dir->d_inode;
3260 save_parent.mnt = NULL;
3261 save_parent.dentry = NULL;
3263 mnt_drop_write(nd->path.mnt);
3270 static int do_tmpfile(int dfd, struct filename *pathname,
3271 struct nameidata *nd, int flags,
3272 const struct open_flags *op,
3273 struct file *file, int *opened)
3275 static const struct qstr name = QSTR_INIT("/", 1);
3276 struct dentry *dentry, *child;
3278 int error = path_lookupat(dfd, pathname,
3279 flags | LOOKUP_DIRECTORY, nd);
3280 if (unlikely(error))
3282 error = mnt_want_write(nd->path.mnt);
3283 if (unlikely(error))
3285 /* we want directory to be writable */
3286 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3289 dentry = nd->path.dentry;
3290 dir = dentry->d_inode;
3291 if (!dir->i_op->tmpfile) {
3292 error = -EOPNOTSUPP;
3295 child = d_alloc(dentry, &name);
3296 if (unlikely(!child)) {
3300 nd->flags &= ~LOOKUP_DIRECTORY;
3301 nd->flags |= op->intent;
3302 dput(nd->path.dentry);
3303 nd->path.dentry = child;
3304 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3307 audit_inode(pathname, nd->path.dentry, 0);
3308 /* Don't check for other permissions, the inode was just created */
3309 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3312 file->f_path.mnt = nd->path.mnt;
3313 error = finish_open(file, nd->path.dentry, NULL, opened);
3316 error = open_check_o_direct(file);
3319 } else if (!(op->open_flag & O_EXCL)) {
3320 struct inode *inode = file_inode(file);
3321 spin_lock(&inode->i_lock);
3322 inode->i_state |= I_LINKABLE;
3323 spin_unlock(&inode->i_lock);
3326 mnt_drop_write(nd->path.mnt);
3328 path_put(&nd->path);
3332 static struct file *path_openat(int dfd, struct filename *pathname,
3333 struct nameidata *nd, const struct open_flags *op, int flags)
3340 file = get_empty_filp();
3344 file->f_flags = op->open_flag;
3346 if (unlikely(file->f_flags & __O_TMPFILE)) {
3347 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3351 s = path_init(dfd, pathname, flags, nd);
3356 while (!(error = link_path_walk(s, nd)) &&
3357 (error = do_last(nd, file, op, &opened, pathname)) > 0) {
3358 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3359 s = trailing_symlink(nd);
3368 if (!(opened & FILE_OPENED)) {
3372 if (unlikely(error)) {
3373 if (error == -EOPENSTALE) {
3374 if (flags & LOOKUP_RCU)
3379 file = ERR_PTR(error);
3384 struct file *do_filp_open(int dfd, struct filename *pathname,
3385 const struct open_flags *op)
3387 struct nameidata nd, *saved_nd = set_nameidata(&nd);
3388 int flags = op->lookup_flags;
3391 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3392 if (unlikely(filp == ERR_PTR(-ECHILD)))
3393 filp = path_openat(dfd, pathname, &nd, op, flags);
3394 if (unlikely(filp == ERR_PTR(-ESTALE)))
3395 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3396 restore_nameidata(saved_nd);
3400 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3401 const char *name, const struct open_flags *op)
3403 struct nameidata nd, *saved_nd;
3405 struct filename *filename;
3406 int flags = op->lookup_flags | LOOKUP_ROOT;
3409 nd.root.dentry = dentry;
3411 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3412 return ERR_PTR(-ELOOP);
3414 filename = getname_kernel(name);
3415 if (unlikely(IS_ERR(filename)))
3416 return ERR_CAST(filename);
3418 saved_nd = set_nameidata(&nd);
3419 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3420 if (unlikely(file == ERR_PTR(-ECHILD)))
3421 file = path_openat(-1, filename, &nd, op, flags);
3422 if (unlikely(file == ERR_PTR(-ESTALE)))
3423 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3424 restore_nameidata(saved_nd);
3429 static struct dentry *filename_create(int dfd, struct filename *name,
3430 struct path *path, unsigned int lookup_flags)
3432 struct dentry *dentry = ERR_PTR(-EEXIST);
3433 struct nameidata nd;
3436 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3439 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3440 * other flags passed in are ignored!
3442 lookup_flags &= LOOKUP_REVAL;
3444 error = filename_parentat(dfd, name, lookup_flags, &nd);
3446 return ERR_PTR(error);
3449 * Yucky last component or no last component at all?
3450 * (foo/., foo/.., /////)
3452 if (nd.last_type != LAST_NORM)
3454 nd.flags &= ~LOOKUP_PARENT;
3455 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3457 /* don't fail immediately if it's r/o, at least try to report other errors */
3458 err2 = mnt_want_write(nd.path.mnt);
3460 * Do the final lookup.
3462 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3463 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3468 if (d_is_positive(dentry))
3472 * Special case - lookup gave negative, but... we had foo/bar/
3473 * From the vfs_mknod() POV we just have a negative dentry -
3474 * all is fine. Let's be bastards - you had / on the end, you've
3475 * been asking for (non-existent) directory. -ENOENT for you.
3477 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3481 if (unlikely(err2)) {
3489 dentry = ERR_PTR(error);
3491 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3493 mnt_drop_write(nd.path.mnt);
3499 struct dentry *kern_path_create(int dfd, const char *pathname,
3500 struct path *path, unsigned int lookup_flags)
3502 struct filename *filename = getname_kernel(pathname);
3505 if (IS_ERR(filename))
3506 return ERR_CAST(filename);
3507 res = filename_create(dfd, filename, path, lookup_flags);
3511 EXPORT_SYMBOL(kern_path_create);
3513 void done_path_create(struct path *path, struct dentry *dentry)
3516 mutex_unlock(&path->dentry->d_inode->i_mutex);
3517 mnt_drop_write(path->mnt);
3520 EXPORT_SYMBOL(done_path_create);
3522 struct dentry *user_path_create(int dfd, const char __user *pathname,
3523 struct path *path, unsigned int lookup_flags)
3525 struct filename *tmp = getname(pathname);
3528 return ERR_CAST(tmp);
3529 res = filename_create(dfd, tmp, path, lookup_flags);
3533 EXPORT_SYMBOL(user_path_create);
3535 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3537 int error = may_create(dir, dentry);
3542 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3545 if (!dir->i_op->mknod)
3548 error = devcgroup_inode_mknod(mode, dev);
3552 error = security_inode_mknod(dir, dentry, mode, dev);
3556 error = dir->i_op->mknod(dir, dentry, mode, dev);
3558 fsnotify_create(dir, dentry);
3561 EXPORT_SYMBOL(vfs_mknod);
3563 static int may_mknod(umode_t mode)
3565 switch (mode & S_IFMT) {
3571 case 0: /* zero mode translates to S_IFREG */
3580 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3583 struct dentry *dentry;
3586 unsigned int lookup_flags = 0;
3588 error = may_mknod(mode);
3592 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3594 return PTR_ERR(dentry);
3596 if (!IS_POSIXACL(path.dentry->d_inode))
3597 mode &= ~current_umask();
3598 error = security_path_mknod(&path, dentry, mode, dev);
3601 switch (mode & S_IFMT) {
3602 case 0: case S_IFREG:
3603 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3605 case S_IFCHR: case S_IFBLK:
3606 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3607 new_decode_dev(dev));
3609 case S_IFIFO: case S_IFSOCK:
3610 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3614 done_path_create(&path, dentry);
3615 if (retry_estale(error, lookup_flags)) {
3616 lookup_flags |= LOOKUP_REVAL;
3622 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3624 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3627 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3629 int error = may_create(dir, dentry);
3630 unsigned max_links = dir->i_sb->s_max_links;
3635 if (!dir->i_op->mkdir)
3638 mode &= (S_IRWXUGO|S_ISVTX);
3639 error = security_inode_mkdir(dir, dentry, mode);
3643 if (max_links && dir->i_nlink >= max_links)
3646 error = dir->i_op->mkdir(dir, dentry, mode);
3648 fsnotify_mkdir(dir, dentry);
3651 EXPORT_SYMBOL(vfs_mkdir);
3653 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3655 struct dentry *dentry;
3658 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3661 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3663 return PTR_ERR(dentry);
3665 if (!IS_POSIXACL(path.dentry->d_inode))
3666 mode &= ~current_umask();
3667 error = security_path_mkdir(&path, dentry, mode);
3669 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3670 done_path_create(&path, dentry);
3671 if (retry_estale(error, lookup_flags)) {
3672 lookup_flags |= LOOKUP_REVAL;
3678 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3680 return sys_mkdirat(AT_FDCWD, pathname, mode);
3684 * The dentry_unhash() helper will try to drop the dentry early: we
3685 * should have a usage count of 1 if we're the only user of this
3686 * dentry, and if that is true (possibly after pruning the dcache),
3687 * then we drop the dentry now.
3689 * A low-level filesystem can, if it choses, legally
3692 * if (!d_unhashed(dentry))
3695 * if it cannot handle the case of removing a directory
3696 * that is still in use by something else..
3698 void dentry_unhash(struct dentry *dentry)
3700 shrink_dcache_parent(dentry);
3701 spin_lock(&dentry->d_lock);
3702 if (dentry->d_lockref.count == 1)
3704 spin_unlock(&dentry->d_lock);
3706 EXPORT_SYMBOL(dentry_unhash);
3708 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3710 int error = may_delete(dir, dentry, 1);
3715 if (!dir->i_op->rmdir)
3719 mutex_lock(&dentry->d_inode->i_mutex);
3722 if (is_local_mountpoint(dentry))
3725 error = security_inode_rmdir(dir, dentry);
3729 shrink_dcache_parent(dentry);
3730 error = dir->i_op->rmdir(dir, dentry);
3734 dentry->d_inode->i_flags |= S_DEAD;
3736 detach_mounts(dentry);
3739 mutex_unlock(&dentry->d_inode->i_mutex);
3745 EXPORT_SYMBOL(vfs_rmdir);
3747 static long do_rmdir(int dfd, const char __user *pathname)
3750 struct filename *name;
3751 struct dentry *dentry;
3755 unsigned int lookup_flags = 0;
3757 name = user_path_parent(dfd, pathname,
3758 &path, &last, &type, lookup_flags);
3760 return PTR_ERR(name);
3774 error = mnt_want_write(path.mnt);
3778 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3779 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3780 error = PTR_ERR(dentry);
3783 if (!dentry->d_inode) {
3787 error = security_path_rmdir(&path, dentry);
3790 error = vfs_rmdir(path.dentry->d_inode, dentry);
3794 mutex_unlock(&path.dentry->d_inode->i_mutex);
3795 mnt_drop_write(path.mnt);
3799 if (retry_estale(error, lookup_flags)) {
3800 lookup_flags |= LOOKUP_REVAL;
3806 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3808 return do_rmdir(AT_FDCWD, pathname);
3812 * vfs_unlink - unlink a filesystem object
3813 * @dir: parent directory
3815 * @delegated_inode: returns victim inode, if the inode is delegated.
3817 * The caller must hold dir->i_mutex.
3819 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3820 * return a reference to the inode in delegated_inode. The caller
3821 * should then break the delegation on that inode and retry. Because
3822 * breaking a delegation may take a long time, the caller should drop
3823 * dir->i_mutex before doing so.
3825 * Alternatively, a caller may pass NULL for delegated_inode. This may
3826 * be appropriate for callers that expect the underlying filesystem not
3827 * to be NFS exported.
3829 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3831 struct inode *target = dentry->d_inode;
3832 int error = may_delete(dir, dentry, 0);
3837 if (!dir->i_op->unlink)
3840 mutex_lock(&target->i_mutex);
3841 if (is_local_mountpoint(dentry))
3844 error = security_inode_unlink(dir, dentry);
3846 error = try_break_deleg(target, delegated_inode);
3849 error = dir->i_op->unlink(dir, dentry);
3852 detach_mounts(dentry);
3857 mutex_unlock(&target->i_mutex);
3859 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3860 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3861 fsnotify_link_count(target);
3867 EXPORT_SYMBOL(vfs_unlink);
3870 * Make sure that the actual truncation of the file will occur outside its
3871 * directory's i_mutex. Truncate can take a long time if there is a lot of
3872 * writeout happening, and we don't want to prevent access to the directory
3873 * while waiting on the I/O.
3875 static long do_unlinkat(int dfd, const char __user *pathname)
3878 struct filename *name;
3879 struct dentry *dentry;
3883 struct inode *inode = NULL;
3884 struct inode *delegated_inode = NULL;
3885 unsigned int lookup_flags = 0;
3887 name = user_path_parent(dfd, pathname,
3888 &path, &last, &type, lookup_flags);
3890 return PTR_ERR(name);
3893 if (type != LAST_NORM)
3896 error = mnt_want_write(path.mnt);
3900 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3901 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3902 error = PTR_ERR(dentry);
3903 if (!IS_ERR(dentry)) {
3904 /* Why not before? Because we want correct error value */
3905 if (last.name[last.len])
3907 inode = dentry->d_inode;
3908 if (d_is_negative(dentry))
3911 error = security_path_unlink(&path, dentry);
3914 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3918 mutex_unlock(&path.dentry->d_inode->i_mutex);
3920 iput(inode); /* truncate the inode here */
3922 if (delegated_inode) {
3923 error = break_deleg_wait(&delegated_inode);
3927 mnt_drop_write(path.mnt);
3931 if (retry_estale(error, lookup_flags)) {
3932 lookup_flags |= LOOKUP_REVAL;
3939 if (d_is_negative(dentry))
3941 else if (d_is_dir(dentry))
3948 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3950 if ((flag & ~AT_REMOVEDIR) != 0)
3953 if (flag & AT_REMOVEDIR)
3954 return do_rmdir(dfd, pathname);
3956 return do_unlinkat(dfd, pathname);
3959 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3961 return do_unlinkat(AT_FDCWD, pathname);
3964 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3966 int error = may_create(dir, dentry);
3971 if (!dir->i_op->symlink)
3974 error = security_inode_symlink(dir, dentry, oldname);
3978 error = dir->i_op->symlink(dir, dentry, oldname);
3980 fsnotify_create(dir, dentry);
3983 EXPORT_SYMBOL(vfs_symlink);
3985 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3986 int, newdfd, const char __user *, newname)
3989 struct filename *from;
3990 struct dentry *dentry;
3992 unsigned int lookup_flags = 0;
3994 from = getname(oldname);
3996 return PTR_ERR(from);
3998 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3999 error = PTR_ERR(dentry);
4003 error = security_path_symlink(&path, dentry, from->name);
4005 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4006 done_path_create(&path, dentry);
4007 if (retry_estale(error, lookup_flags)) {
4008 lookup_flags |= LOOKUP_REVAL;
4016 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4018 return sys_symlinkat(oldname, AT_FDCWD, newname);
4022 * vfs_link - create a new link
4023 * @old_dentry: object to be linked
4025 * @new_dentry: where to create the new link
4026 * @delegated_inode: returns inode needing a delegation break
4028 * The caller must hold dir->i_mutex
4030 * If vfs_link discovers a delegation on the to-be-linked file in need
4031 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4032 * inode in delegated_inode. The caller should then break the delegation
4033 * and retry. Because breaking a delegation may take a long time, the
4034 * caller should drop the i_mutex before doing so.
4036 * Alternatively, a caller may pass NULL for delegated_inode. This may
4037 * be appropriate for callers that expect the underlying filesystem not
4038 * to be NFS exported.
4040 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4042 struct inode *inode = old_dentry->d_inode;
4043 unsigned max_links = dir->i_sb->s_max_links;
4049 error = may_create(dir, new_dentry);
4053 if (dir->i_sb != inode->i_sb)
4057 * A link to an append-only or immutable file cannot be created.
4059 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4061 if (!dir->i_op->link)
4063 if (S_ISDIR(inode->i_mode))
4066 error = security_inode_link(old_dentry, dir, new_dentry);
4070 mutex_lock(&inode->i_mutex);
4071 /* Make sure we don't allow creating hardlink to an unlinked file */
4072 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4074 else if (max_links && inode->i_nlink >= max_links)
4077 error = try_break_deleg(inode, delegated_inode);
4079 error = dir->i_op->link(old_dentry, dir, new_dentry);
4082 if (!error && (inode->i_state & I_LINKABLE)) {
4083 spin_lock(&inode->i_lock);
4084 inode->i_state &= ~I_LINKABLE;
4085 spin_unlock(&inode->i_lock);
4087 mutex_unlock(&inode->i_mutex);
4089 fsnotify_link(dir, inode, new_dentry);
4092 EXPORT_SYMBOL(vfs_link);
4095 * Hardlinks are often used in delicate situations. We avoid
4096 * security-related surprises by not following symlinks on the
4099 * We don't follow them on the oldname either to be compatible
4100 * with linux 2.0, and to avoid hard-linking to directories
4101 * and other special files. --ADM
4103 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4104 int, newdfd, const char __user *, newname, int, flags)
4106 struct dentry *new_dentry;
4107 struct path old_path, new_path;
4108 struct inode *delegated_inode = NULL;
4112 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4115 * To use null names we require CAP_DAC_READ_SEARCH
4116 * This ensures that not everyone will be able to create
4117 * handlink using the passed filedescriptor.
4119 if (flags & AT_EMPTY_PATH) {
4120 if (!capable(CAP_DAC_READ_SEARCH))
4125 if (flags & AT_SYMLINK_FOLLOW)
4126 how |= LOOKUP_FOLLOW;
4128 error = user_path_at(olddfd, oldname, how, &old_path);
4132 new_dentry = user_path_create(newdfd, newname, &new_path,
4133 (how & LOOKUP_REVAL));
4134 error = PTR_ERR(new_dentry);
4135 if (IS_ERR(new_dentry))
4139 if (old_path.mnt != new_path.mnt)
4141 error = may_linkat(&old_path);
4142 if (unlikely(error))
4144 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4147 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4149 done_path_create(&new_path, new_dentry);
4150 if (delegated_inode) {
4151 error = break_deleg_wait(&delegated_inode);
4153 path_put(&old_path);
4157 if (retry_estale(error, how)) {
4158 path_put(&old_path);
4159 how |= LOOKUP_REVAL;
4163 path_put(&old_path);
4168 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4170 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4174 * vfs_rename - rename a filesystem object
4175 * @old_dir: parent of source
4176 * @old_dentry: source
4177 * @new_dir: parent of destination
4178 * @new_dentry: destination
4179 * @delegated_inode: returns an inode needing a delegation break
4180 * @flags: rename flags
4182 * The caller must hold multiple mutexes--see lock_rename()).
4184 * If vfs_rename discovers a delegation in need of breaking at either
4185 * the source or destination, it will return -EWOULDBLOCK and return a
4186 * reference to the inode in delegated_inode. The caller should then
4187 * break the delegation and retry. Because breaking a delegation may
4188 * take a long time, the caller should drop all locks before doing
4191 * Alternatively, a caller may pass NULL for delegated_inode. This may
4192 * be appropriate for callers that expect the underlying filesystem not
4193 * to be NFS exported.
4195 * The worst of all namespace operations - renaming directory. "Perverted"
4196 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4198 * a) we can get into loop creation.
4199 * b) race potential - two innocent renames can create a loop together.
4200 * That's where 4.4 screws up. Current fix: serialization on
4201 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4203 * c) we have to lock _four_ objects - parents and victim (if it exists),
4204 * and source (if it is not a directory).
4205 * And that - after we got ->i_mutex on parents (until then we don't know
4206 * whether the target exists). Solution: try to be smart with locking
4207 * order for inodes. We rely on the fact that tree topology may change
4208 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4209 * move will be locked. Thus we can rank directories by the tree
4210 * (ancestors first) and rank all non-directories after them.
4211 * That works since everybody except rename does "lock parent, lookup,
4212 * lock child" and rename is under ->s_vfs_rename_mutex.
4213 * HOWEVER, it relies on the assumption that any object with ->lookup()
4214 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4215 * we'd better make sure that there's no link(2) for them.
4216 * d) conversion from fhandle to dentry may come in the wrong moment - when
4217 * we are removing the target. Solution: we will have to grab ->i_mutex
4218 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4219 * ->i_mutex on parents, which works but leads to some truly excessive
4222 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4223 struct inode *new_dir, struct dentry *new_dentry,
4224 struct inode **delegated_inode, unsigned int flags)
4227 bool is_dir = d_is_dir(old_dentry);
4228 const unsigned char *old_name;
4229 struct inode *source = old_dentry->d_inode;
4230 struct inode *target = new_dentry->d_inode;
4231 bool new_is_dir = false;
4232 unsigned max_links = new_dir->i_sb->s_max_links;
4234 if (source == target)
4237 error = may_delete(old_dir, old_dentry, is_dir);
4242 error = may_create(new_dir, new_dentry);
4244 new_is_dir = d_is_dir(new_dentry);
4246 if (!(flags & RENAME_EXCHANGE))
4247 error = may_delete(new_dir, new_dentry, is_dir);
4249 error = may_delete(new_dir, new_dentry, new_is_dir);
4254 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4257 if (flags && !old_dir->i_op->rename2)
4261 * If we are going to change the parent - check write permissions,
4262 * we'll need to flip '..'.
4264 if (new_dir != old_dir) {
4266 error = inode_permission(source, MAY_WRITE);
4270 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4271 error = inode_permission(target, MAY_WRITE);
4277 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4282 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4284 if (!is_dir || (flags & RENAME_EXCHANGE))
4285 lock_two_nondirectories(source, target);
4287 mutex_lock(&target->i_mutex);
4290 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4293 if (max_links && new_dir != old_dir) {
4295 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4297 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4298 old_dir->i_nlink >= max_links)
4301 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4302 shrink_dcache_parent(new_dentry);
4304 error = try_break_deleg(source, delegated_inode);
4308 if (target && !new_is_dir) {
4309 error = try_break_deleg(target, delegated_inode);
4313 if (!old_dir->i_op->rename2) {
4314 error = old_dir->i_op->rename(old_dir, old_dentry,
4315 new_dir, new_dentry);
4317 WARN_ON(old_dir->i_op->rename != NULL);
4318 error = old_dir->i_op->rename2(old_dir, old_dentry,
4319 new_dir, new_dentry, flags);
4324 if (!(flags & RENAME_EXCHANGE) && target) {
4326 target->i_flags |= S_DEAD;
4327 dont_mount(new_dentry);
4328 detach_mounts(new_dentry);
4330 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4331 if (!(flags & RENAME_EXCHANGE))
4332 d_move(old_dentry, new_dentry);
4334 d_exchange(old_dentry, new_dentry);
4337 if (!is_dir || (flags & RENAME_EXCHANGE))
4338 unlock_two_nondirectories(source, target);
4340 mutex_unlock(&target->i_mutex);
4343 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4344 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4345 if (flags & RENAME_EXCHANGE) {
4346 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4347 new_is_dir, NULL, new_dentry);
4350 fsnotify_oldname_free(old_name);
4354 EXPORT_SYMBOL(vfs_rename);
4356 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4357 int, newdfd, const char __user *, newname, unsigned int, flags)
4359 struct dentry *old_dentry, *new_dentry;
4360 struct dentry *trap;
4361 struct path old_path, new_path;
4362 struct qstr old_last, new_last;
4363 int old_type, new_type;
4364 struct inode *delegated_inode = NULL;
4365 struct filename *from;
4366 struct filename *to;
4367 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4368 bool should_retry = false;
4371 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4374 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4375 (flags & RENAME_EXCHANGE))
4378 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4381 if (flags & RENAME_EXCHANGE)
4385 from = user_path_parent(olddfd, oldname,
4386 &old_path, &old_last, &old_type, lookup_flags);
4388 error = PTR_ERR(from);
4392 to = user_path_parent(newdfd, newname,
4393 &new_path, &new_last, &new_type, lookup_flags);
4395 error = PTR_ERR(to);
4400 if (old_path.mnt != new_path.mnt)
4404 if (old_type != LAST_NORM)
4407 if (flags & RENAME_NOREPLACE)
4409 if (new_type != LAST_NORM)
4412 error = mnt_want_write(old_path.mnt);
4417 trap = lock_rename(new_path.dentry, old_path.dentry);
4419 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4420 error = PTR_ERR(old_dentry);
4421 if (IS_ERR(old_dentry))
4423 /* source must exist */
4425 if (d_is_negative(old_dentry))
4427 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4428 error = PTR_ERR(new_dentry);
4429 if (IS_ERR(new_dentry))
4432 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4434 if (flags & RENAME_EXCHANGE) {
4436 if (d_is_negative(new_dentry))
4439 if (!d_is_dir(new_dentry)) {
4441 if (new_last.name[new_last.len])
4445 /* unless the source is a directory trailing slashes give -ENOTDIR */
4446 if (!d_is_dir(old_dentry)) {
4448 if (old_last.name[old_last.len])
4450 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4453 /* source should not be ancestor of target */
4455 if (old_dentry == trap)
4457 /* target should not be an ancestor of source */
4458 if (!(flags & RENAME_EXCHANGE))
4460 if (new_dentry == trap)
4463 error = security_path_rename(&old_path, old_dentry,
4464 &new_path, new_dentry, flags);
4467 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4468 new_path.dentry->d_inode, new_dentry,
4469 &delegated_inode, flags);
4475 unlock_rename(new_path.dentry, old_path.dentry);
4476 if (delegated_inode) {
4477 error = break_deleg_wait(&delegated_inode);
4481 mnt_drop_write(old_path.mnt);
4483 if (retry_estale(error, lookup_flags))
4484 should_retry = true;
4485 path_put(&new_path);
4488 path_put(&old_path);
4491 should_retry = false;
4492 lookup_flags |= LOOKUP_REVAL;
4499 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4500 int, newdfd, const char __user *, newname)
4502 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4505 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4507 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4510 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4512 int error = may_create(dir, dentry);
4516 if (!dir->i_op->mknod)
4519 return dir->i_op->mknod(dir, dentry,
4520 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4522 EXPORT_SYMBOL(vfs_whiteout);
4524 int readlink_copy(char __user *buffer, int buflen, const char *link)
4526 int len = PTR_ERR(link);
4531 if (len > (unsigned) buflen)
4533 if (copy_to_user(buffer, link, len))
4538 EXPORT_SYMBOL(readlink_copy);
4541 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4542 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4543 * using) it for any given inode is up to filesystem.
4545 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4548 struct inode *inode = d_inode(dentry);
4549 const char *link = inode->i_link;
4553 link = inode->i_op->follow_link(dentry, &cookie);
4555 return PTR_ERR(link);
4557 res = readlink_copy(buffer, buflen, link);
4558 if (inode->i_op->put_link)
4559 inode->i_op->put_link(inode, cookie);
4562 EXPORT_SYMBOL(generic_readlink);
4564 /* get the link contents into pagecache */
4565 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4569 struct address_space *mapping = dentry->d_inode->i_mapping;
4570 page = read_mapping_page(mapping, 0, NULL);
4575 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4579 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4581 struct page *page = NULL;
4582 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4585 page_cache_release(page);
4589 EXPORT_SYMBOL(page_readlink);
4591 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4593 struct page *page = NULL;
4594 char *res = page_getlink(dentry, &page);
4599 EXPORT_SYMBOL(page_follow_link_light);
4601 void page_put_link(struct inode *unused, void *cookie)
4603 struct page *page = cookie;
4605 page_cache_release(page);
4607 EXPORT_SYMBOL(page_put_link);
4610 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4612 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4614 struct address_space *mapping = inode->i_mapping;
4619 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4621 flags |= AOP_FLAG_NOFS;
4624 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4625 flags, &page, &fsdata);
4629 kaddr = kmap_atomic(page);
4630 memcpy(kaddr, symname, len-1);
4631 kunmap_atomic(kaddr);
4633 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4640 mark_inode_dirty(inode);
4645 EXPORT_SYMBOL(__page_symlink);
4647 int page_symlink(struct inode *inode, const char *symname, int len)
4649 return __page_symlink(inode, symname, len,
4650 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4652 EXPORT_SYMBOL(page_symlink);
4654 const struct inode_operations page_symlink_inode_operations = {
4655 .readlink = generic_readlink,
4656 .follow_link = page_follow_link_light,
4657 .put_link = page_put_link,
4659 EXPORT_SYMBOL(page_symlink_inode_operations);